Broadcom Slides to $393.02 as the Cheapest AI Asset on the Board — Bookings Run 3x Shipments and the Market Charges for Concentration

Broadcom trades at $393.02, down 3.40%, holding a session range of $386.74 to $399.00 on 16.32 million shares against average daily volume of 23.12 million. Market capitalization sits at $1.88 trillion. The stock closed at $394.28 on Wednesday.

The 52-week high printed $495.00 on June 3, 2026. The 52-week low was $273.00 on July 23, 2025.

At $393.02, AVGO sits 20.6% below the high it set six weeks ago and 43.9% above the low it set twelve months ago.

The sector took it down. The Technology Equipment complex fell 2.33% and Broadcom underperformed the industry. Micron dropped 4.60%. SanDisk fell 7.81%. Nvidia lost 2.30%. A gauge of semiconductor firms shed 3% while the Nasdaq Composite fell 0.82% to 26,054.38.

The decline is largely a broader cooling in the semiconductor industry as market participants reassess the sustainability of high-valuation growth trajectories, against a backdrop of renewed rate concerns where economic data has prompted a sell-off in duration-sensitive assets. As a bellwether for the AI infrastructure trade, Broadcom carries heightened sensitivity to shifts in institutional risk appetite when macro indicators point to a prolonged period of elevated borrowing costs.

The 10-year Treasury yield ripped to 4.60% today, approaching its 4.62% two-month high.

The thesis is uncomfortable and specific. Broadcom is the only AI infrastructure name whose bookings run three times its shipments and whose stock is up 13% year to date. The reason it is cheap is that Hock Tan refused to raise a $100 billion target and got punished 15% for it — and that its three largest customers are all building their own chips. At 24.37x forward earnings against a $30 billion booking quarter, this is where the AI trade is priced for failure.

The company announced a long-term strategic commitment with Standard Chartered today to accelerate the bank's global infrastructure modernization through a private cloud foundation. The stock fell 3.40%.

The fundamentals underneath are extreme. Trailing twelve-month revenue runs $75.465 billion at a 68.35% gross margin and a 38.85% net margin. EBITDA sits at $42.063 billion. Return on equity is 37.28%. Beta reads 1.47 across 4.76 billion shares outstanding.

Up 13% Year to Date While the AI Trade Ran

The performance gap is the entire setup and it is the first thing worth stating.

Broadcom stock has risen only 13% year to date, underperforming several chip stocks amid concerns about rising competition from Taiwan-based MediaTek.

Thirteen percent. On a company whose AI semiconductor revenue grew 143% year over year.

Compare the field. TSMC is up almost 40% year to date at 411.20 after posting record Q2 revenue of $40.2 billion and raising 2026 capex to $60-64 billion. Apple hit an all-time high of 327.50 on Wednesday. Amazon at $256.87 has gained 14.22% over twelve months on a cloud business growing 28%.

Broadcom's AI business grew 143% and the stock delivered 13%.

That divergence has a specific cause and a specific date. The stock was down about 15% on the Thursday after Q2 earnings on disappointment that CEO Hock Tan did not raise the company's full-year target of $100 billion in AI chip sales.

Fifteen percent. For not raising a number he had already given.

Run the arithmetic on what that cost. AVGO at $495.00 on June 3 — the day it reported — against $393.02 today is a $101.98 decline, or 20.6%. At 4.76 billion shares, that is $485 billion of market capitalization removed from a company that simultaneously reported record revenue, record operating profit, record free cash flow and $30 billion of AI bookings.

Half a trillion dollars, for declining to raise guidance.

The tape is doing the same thing today that it did to TSMC. TSMC printed a record quarter, raised capex to $60-64 billion, lifted revenue growth guidance above 40% and got sold 1.97%. GE Aerospace raised full-year guidance $700 million at the midpoint and fell 4.00%.

Good news has stopped paying inside the AI trade. Broadcom is the purest expression of that, because it has been living under the verdict for six weeks while everyone else is discovering it today.

The Magnificent 7 are averaging roughly +3.7% year to date. Broadcom at 13% is beating them and it does not feel like it.

The $30 Billion Booking Quarter Against $10.8 Billion Shipped

Here is the number that should not coexist with a 20.6% drawdown.

During the second quarter, bookings for AI semiconductors were over $30 billion against the $10.8 billion Broadcom shipped. Bookings now exceed three times quarterly shipments, reflecting structural demand acceleration and a growing backlog.

Three to one. Orders to shipments. In a single quarter.

That ratio is the single most bullish operating metric in the semiconductor complex and it is being ignored because it does not convert to revenue this quarter.

Tan was explicit about why: the bookings that are coming are not for immediate delivery.

That sentence is the entire disconnect between this company's fundamentals and its stock price. A $30 billion order book with delivery in 2027 and 2028 is worth less to a market discounting at 4.60% than $10.8 billion of revenue booked today — which is precisely the duration problem that has repriced every AI infrastructure name this quarter.

The backlog history shows how fast it compounds. In December, Broadcom disclosed an AI backlog over the next 18 months of $73 billion. The market spread that linearly over six quarters, but backlogs are front-loaded — roughly 80% or more of that $73 billion gets delivered in the first four quarters.

Analysts asked directly whether the 18-month backlog covering the second half of fiscal 2026 through all of fiscal 2027 now sits at $200 billion or better, citing the strength of all programs, the broadening customer base, accelerating year-over-year AI shipments and the multi-gigawatt partnerships set to start firing next year.

Tan called it a very complicated set of number questions and did not answer.

That non-answer cost 15%.

Management highlighted unprecedented order visibility extending into 2028 from hyperscaler and large language model customers, with multi-year agreements underpinning long-term expectations of $100 billion-plus in annual AI semiconductor revenue starting in fiscal 2027.

On supply, Tan said the company is comfortable for 2026 and 2027 and is working on securing supply for 2028 and 2029, having met incremental customer demand and expecting to continue.

Booked through 2028. Supply secured through 2027. Stock down 20.6%.

AI at $10.8 Billion Is Now Half the Company

The transformation is complete and the numbers are stark.

Q2 semiconductor revenue from AI hit a record $10.8 billion, up 143% year over year and above forecast, driven by increasing demand for custom AI accelerators and AI networking. That figure represented roughly 49% of total revenue.

Almost half of every dollar Broadcom earned came from AI chips.

Total revenue reached a record $22.2 billion, up 48% year over year. Semiconductor revenue hit a record $15 billion, up 79%. Operating margin printed a record 67% and adjusted EBITDA a record 69% of revenue — both above guidance. Non-GAAP EPS landed at $2.44 against roughly $2.32 expected.

The acceleration across four quarters is the part that gets lost. Q1 fiscal 2026 AI revenue ran $8.4 billion, up 106% year over year, on consolidated revenue of $19.3 billion that grew 29%. Adjusted EBITDA hit a record $13.1 billion at 68% of revenue. Management guided Q2 AI to $10.7 billion and delivered $10.8 billion.

From $8.4 billion to $10.8 billion in one quarter. Guidance beaten. Growth accelerating from 106% to 143%.

Networking represented almost 40% of Q2 AI revenue — the piece nobody models. Broadcom does not just sell the accelerator. It sells the switching fabric that connects them, which is a franchise Nvidia has spent two years trying to buy its way into.

Demand for XPUs and networking is simply insatiable. That is Tan's word, on the record, on the call.

The model is what makes it durable. Broadcom does not sell off-the-shelf GPUs the way Nvidia does. It designs custom AI accelerators — XPUs — plus networking for specific hyperscaler clients. That creates sticky multi-year relationships, because a custom chip is co-engineered with the customer and switching costs are high.

That contrasts directly with the commercial-GPU approach at Nvidia and the broader lineup at AMD.

Q3 guidance calls for AI semiconductor revenue growing over 200% year over year to $16.0 billion, with total revenue guided to about $29.4 billion.

Sixteen billion of AI revenue in a single quarter. That is more than the company's entire AI business generated in fiscal 2025.

Gross Margin at 77.1% Falling to 74% — the Mix Problem

This is the part the bulls skip and it deserves the airing.

Q2 gross margin declined 230 basis points year over year to 77.1% due to the increasing proportion of semiconductor products in the mix. The company faces pressure on semiconductor margins from the lower margins of ASICs and TPUs. Consolidated gross margin is expected to be approximately 74% in Q3.

From 77.1% to 74% in one quarter. A 310 basis point compression, guided.

Run what that does. At $29.4 billion of Q3 revenue, each 100 basis points of gross margin is $294 million. A 310 basis point decline removes $911 million of gross profit against a revenue base growing 32% sequentially.

The rapid growth in AI semiconductor revenue is causing the decline in consolidated gross margin. That is the mechanism stated plainly, and it is structural rather than cyclical.

Management's defense is technically correct and worth registering. The shift toward AI semiconductors elevated the product mix and resulted in a gross margin decline which was not due to margin erosion at the segment level.

Read that carefully. No individual product got less profitable. The company is simply selling more of the lower-margin thing and less of the higher-margin thing — because the lower-margin thing is growing 143% and the higher-margin thing is growing 9%.

That is the trade Broadcom has made. Custom ASICs and TPUs carry structurally lower margins than infrastructure software because the customer co-designs them, knows the cost stack and negotiates accordingly. Google does not pay Nvidia margins for a chip Google helped engineer.

The offset is operating leverage and it is genuinely impressive. The company achieved substantial operating income and free cash flow growth without a corresponding increase in operating expenses. Operating margin hit a record 67% while gross margin fell 230 basis points.

Revenue scaling massively while operating and EBITDA margins remain strong and stable. That is Tan's framing and the numbers support it.

But the direction is one way. As AI grows from 49% of revenue toward 60% and 70%, consolidated gross margin keeps falling toward the ASIC blend. At 74% in Q3, the market is watching a 68.35% trailing figure that has further to compress.

Trailing net margin sits at 38.85%. That is the number to watch through fiscal 2027.

The $56 Billion Number Nobody Could Square

The guidance confusion is what triggered the selloff and it is worth reconstructing precisely.

Broadcom forecasts $56 billion of AI semiconductor revenue for fiscal 2026 — up approximately 180% from fiscal 2025 — and expects AI semiconductor revenue to double in the second half.

The first half shipped roughly $19 billion in total AI revenue. Q1 delivered $8.4 billion and Q2 delivered $10.8 billion, for $19.2 billion.

Tan's math: take $19 billion, 2x it in the second half, and you get pretty much in the range of $56 billion.

The analyst pushback was immediate and it is the crux. If AI grows 2x second half over first half, that would put AI revenues over $60 billion with sequential growth in fiscal Q4 — but the $56 billion number implies only about 1.5x half-over-half growth, with Q4 AI actually declining sequentially.

Someone asked the CEO to square his own guidance on a live call and he called it a very complicated set of number questions.

Run it independently. $56 billion full year minus $19.2 billion first half leaves $36.8 billion for the second half — 1.92x the first half, not 2x. Q3 is guided to $16.0 billion, implying Q4 at $20.8 billion. That is sequential growth, not decline.

The numbers work. The communication did not.

That gap between a guide that arithmetically holds and a CEO who could not articulate it in real time is what produced a 15% single-day decline in a $2 trillion company.

The forward number is the one that matters and Tan reiterated rather than raised it: AI semiconductor revenue in excess of $100 billion for fiscal 2027, with momentum expected to continue.

Read the sequence. December: $73 billion 18-month backlog. Q2: $30 billion of bookings in one quarter against $10.8 billion shipped. Fiscal 2027: in excess of $100 billion.

If Q2's $30 billion booking rate held for four quarters, that is $120 billion of orders annually. The company is guiding fiscal 2027 to "in excess of $100 billion."

Tan is guiding below his own order book, and the market punished him for not guiding above it.

Apple's $30 Billion Through 2031

The deal that should have re-rated this stock landed on July 8 and did almost nothing.

Apple and Broadcom announced a renewed chip supply agreement through 2031. The agreement exceeds $30 billion in chips to Apple and includes wireless products like Broadcom's FBAR filters as well as custom chips, likely for artificial intelligence.

Five years. Thirty billion dollars. From the largest consumer hardware company on earth.

The stock rode the $30 billion Apple chip deal to defy the tech sell-off on the day. It has since given it back and more.

The strategic read is the part that is underpriced. Apple has spent a decade systematically removing suppliers from its bill of materials — Intel modems, Imagination graphics, Dialog power management, Qualcomm basebands. Broadcom was the name most consistently on the list of companies Apple was expected to design out.

Instead Apple extended through 2031 and added custom AI silicon.

The Evercore read was that the agreement is a strategic positive for Apple. The read for Broadcom is larger: the customer that eliminates suppliers just committed to five more years and expanded the scope.

Apple hit an all-time high of 327.50 on Wednesday, up 12.64 or 4.01%. The company on the other side of a $30 billion contract is ripping. The supplier is down 3.40%.

The FBAR filter franchise is the boring half and it is the reliable half. Every iPhone carries Broadcom's radio frequency filters. That is a decade-long annuity attached to the highest-volume premium hardware program in existence, and it has nothing to do with AI capex cycles or hyperscaler budgets.

The custom chips likely for AI are the new piece. Apple building AI silicon with Broadcom means a fourth major custom accelerator customer alongside Google, Anthropic and OpenAI — and it means Apple's on-device and data center AI roadmap runs through Palo Alto.

Broadcom is best-of-breed in custom AI accelerators, boasting the largest customer in Google and layering new customers in like Anthropic and OpenAI.

Add Apple through 2031 and the customer list is the entire frontier.

Google Is the Largest Customer and MediaTek Just Took a Piece

Here is the risk that has cost this stock 20.6% and it is genuine.

Broadcom's chip business bears significant customer concentration, with a small handful of large AI customers driving the bulk of revenue and future growth. Google is the largest customer.

MediaTek is encroaching on Broadcom's custom ASIC business via a deal with Google.

That is the single sentence behind the underperformance. AVGO has risen only 13% year to date amid concerns about rising competition from Taiwan-based MediaTek.

Read the exposure. Google's TPU program is the anchor of Broadcom's custom accelerator franchise. It is the design win that created the category, the reference customer that brought Meta, Anthropic and OpenAI to the door, and the largest single revenue contributor. If Google dual-sources to MediaTek, the concentration that made this business also breaks it.

Morgan Stanley reaffirmed a Buy on July 14, arguing Broadcom remains a core artificial intelligence winner even as MediaTek encroaches on its custom ASIC business via the Google deal.

That framing is the bull case and it is defensible. A custom chip is co-engineered over years. Switching a TPU generation to a new partner costs Google time it does not have in a compute race. MediaTek taking a socket is not MediaTek taking the program.

The counterweight is that it establishes a precedent. Google spent a decade single-sourcing custom silicon to Broadcom. It no longer does. Every subsequent negotiation happens with a credible alternative in the room, and that shows up in gross margin before it shows up in revenue.

Which is exactly what the 77.1% to 74% guide describes.

The regulatory pressure compounds it. The Cloud Infrastructure Services Providers in Europe joined four other trade groups urging EU antitrust regulators to suspend some business practices of Broadcom. The company is also facing scrutiny amid a high-profile patent infringement investigation involving Samsung and Netlist.

Those are the friction costs of owning a monopoly position. They are also what a monopoly position looks like from the outside.

Customer concentration is the thesis and the risk. It is the same sentence.

OpenAI and Anthropic Are Building Their Own

The second-order risk is more interesting than the MediaTek story and it is being read exactly backwards.

OpenAI revealed its first custom chip in late June. Anthropic is in talks to develop its own chip from the ground up, partnering with a key manufacturer.

The reflexive read is that Broadcom's customers are becoming competitors. The correct read is that Broadcom's customers are becoming customers.

Broadcom has secured long-term agreements with major tech companies including Google, Anthropic and OpenAI, ensuring sustained business growth. It is layering new customers in like Anthropic and OpenAI.

Nobody builds a frontier accelerator alone. OpenAI revealing a custom chip means OpenAI hired someone to co-engineer, tape out, package and network it. That someone has a $30 billion booking quarter and 3x order coverage.

Custom silicon is not a threat to the custom silicon company. It is the product.

The company that should be worried is Nvidia, and the tape is pricing it that way in miniature — NVDA fell 2.30% today against Broadcom's 3.40%, but Nvidia's franchise is the merchant GPU that every one of these custom programs is designed to displace.

The multi-gigawatt partnerships Tan articulated are set to start firing next year. That is the OpenAI and Anthropic pipeline arriving in fiscal 2027 — precisely when the $100 billion-plus AI target begins.

The order visibility extends into 2028 from hyperscaler and large language model customers. LLM customers are new. They did not exist in the backlog eighteen months ago.

The competitive dynamic to watch is the one nobody frames correctly: Broadcom's risk is not that customers build their own chips. It is that customers build their own chips with someone else — which is what the MediaTek-Google deal establishes as possible.

Anthropic is in talks with a key manufacturer. The identity of that manufacturer is worth more to this stock than any macro print.

Chips Only: The Strategy Shift Nobody Priced

Buried in the Q2 call is a strategic reversal that received almost no attention and may matter more than the guidance.

Tan said the company would offer chips only, instead of the complete integrated AI systems Broadcom had previously said it would be providing to its customers.

That is a retreat from the rack. And it is probably correct.

Read what it means. Broadcom had been positioning to sell full integrated AI systems — accelerators, networking, the whole assembly — competing directly with Nvidia's rack-scale approach. Tan just pulled back to selling components.

The bear read is that Broadcom lost a strategic ambition and shrank its addressable market at exactly the wrong moment.

The bull read is that Tan looked at the margin math and refused. Integrated systems carry hardware assembly economics. Broadcom runs a 68.35% trailing gross margin and a 67% operating margin. Adding rack integration would crater both to fund revenue that does not improve returns.

That is the same discipline that produced the refusal to raise the $100 billion target. Hock Tan has run this company for two decades on the principle that margin beats revenue, and the market has rewarded it with a $1.88 trillion valuation and a 37.28% return on equity.

He just applied it twice in one quarter and lost 20.6%.

The software segment is where that discipline lives. Q2 software revenue rose 9% year over year — a small shortfall — and is expected to grow 31% in Q3. That segment is the VMware franchise, and it carries the margins that make the consolidated 77.1% possible.

Today's Standard Chartered announcement is exactly that business at work: a long-term strategic commitment to accelerate the bank's global infrastructure modernization by establishing a private cloud foundation. Palo Alto, London and Singapore.

That is a multi-year enterprise software annuity, announced on a day the stock fell 3.40% because Micron dropped 4.60%.

The software business is 51% of revenue and 100% of the margin cushion. Nobody is watching it.

24.37x Forward Against 64.78x Trailing

The valuation is the argument and the two multiples tell opposite stories.

AVGO carries a trailing P/E of 64.78 on TTM EPS of $6.00. Forward P/E on next twelve months sits at 24.37.

That gap — 64.78x to 24.37x — is the market pricing earnings roughly 166% higher over the next year.

Run it. Trailing EPS of $6.00 against a forward multiple of 24.37 at $393.02 implies forward EPS near $16.13. Q2 delivered $2.44 non-GAAP. Q3 is guided to about $29.4 billion of revenue against $22.2 billion in Q2 — 32% sequential growth.

The forward multiple is not expensive. It is what a company growing AI revenue 143% and total revenue 48% should trade at if the growth is real.

The trailing multiple is what you pay if it is not.

Compare the field. Amazon trades at 12x EV/EBITDA with free cash flow of $1.2 billion. GE Aerospace carried 48.01x forward into a print that raised guidance $700 million and fell 4.00%. TSMC trades at 40% year-to-date gains with AI at 61% of revenue.

Broadcom at 24.37x forward, with 3x booking coverage, order visibility into 2028, a 67% operating margin and $42.063 billion of trailing EBITDA, is the cheapest large-cap AI infrastructure asset on the board.

Morningstar has it trading at a premium. The market has it down 20.6% from the June high.

The dividend is the tell that this is not a speculative asset. AVGO pays $2.60 annually at $0.65 per quarter for a 0.65% yield, with the last ex-date on June 22. Debt-to-equity sits at 74.02%.

A dividend-paying, cash-generative, 38.85% net margin business trading at 24.37x forward is not an AI bet. It is an industrial with an AI attachment.

The upside case is on the record. Infrastructure Capital's Jay Hatfield sees upside to $600 — 52.7% above spot and 21.2% above the all-time high.

The insider flow cuts against it. Chief Legal Officer Mark Brazeal sold 25,000 shares on July 10 for $10.03 million. Cypress Funds' Steven Baum cut his Broadcom stake.

Next earnings land around September 2.

What Has to Break

Map the bull case. The $30 billion booking quarter converts — bookings running 3x shipments cannot stay theoretical past fiscal 2027, when the $100 billion-plus AI target begins. Q3 delivers $16.0 billion of AI revenue at over 200% growth against a $29.4 billion total guide. The multi-gigawatt partnerships with OpenAI and Anthropic start firing next year as Tan indicated. Apple's $30 billion through 2031 expands into custom AI silicon. The 18-month backlog through fiscal 2027 confirms nearer $200 billion than $100 billion. Gross margin stabilizes near 74% as software grows 31% and offsets the ASIC mix.

On that path, 24.37x forward on a business compounding AI at triple digits is a mispricing, and Hatfield's $600 is the reference rather than the fantasy.

Map the bear case. MediaTek's Google socket becomes MediaTek's Google program, and the largest customer dual-sources permanently. Gross margin keeps compressing below 74% as ASICs and TPUs take share from software. Customer concentration bites — a small handful of AI buyers driving the bulk of revenue means one budget cut is a guidance reset. The EU antitrust groups get traction. The 10-year holds 4.60% and every dollar of 2028 backlog discounts harder. Tan reiterates $100 billion again in September and the market takes another 15%.

On that path, $386.74 fails and the $273.00 low from twelve months ago stops being unthinkable.

The base case sits where the stock is. AVGO chops between $378 and $399 into the September 2 print, with the backlog providing a floor and the customer concentration providing a ceiling.

What makes Broadcom different from every other name in this selloff: TSMC got sold for raising capex. GE got sold for raising guidance. Broadcom got sold for refusing to raise a target it had already given — which is the only one of the three that reflects discipline rather than disappointment.

Hock Tan told the market his AI business would exceed $100 billion in fiscal 2027 and declined to say $120 billion. The stock lost $485 billion.

Bookings are $30 billion a quarter. Shipments are $10.8 billion. The difference is the whole trade.

That's TradingNEWS